Saturday November 21, 2009 4:14 PM AEST

DRAM error rates higher than thought

  • Email a Friend
  • Print Page
DRAM error rates higher than thought
By Justin Robinson
Oct 7, 2009 | 4 Comments
Tags: DRAM | error | rates | higher | than | thought

Even ECC is affected.

Solid state memory, such as the DRAM modules running in the very computer you're likely reading this on, is typically thought of as pretty flawless.

While errors are commonplace for processors (making them many times per second), memory is never thought of first when files are corrupted - but a study by a Google employee seems to suggest otherwise.

Taking a survey of the thousands of servers under their control, Google tracked thousands of DRAM modules across every type from DDR1 to DDR2 in densities ranging from the smallest to the largest.

Surprisingly they found that it didn't matter what type of memory was being used at the time; while 8% of modules have errors, this averages out to a whopping 3,751 errors per stick every year.

The study seems to narrow the problem down to the motherboards that the DRAM is running in rather than the sticks themselves, but if a server experiences an error it is incredibly likely to experience another one in that same year.

It's definitely something to think about for those running mission-critical applications that run in memory exclusively, considering that standard DIMMs we run don't have even basic error checking.

Head over to ZDnet to read a little more about the problem.

 

 
 
The latest issue is on sale now!

Want to learn all about Diablo III? Want to find out what the best Solid State Drive is on the market today, and how to look after it? Want to catch up on the latest hardware, games and in depth tech from Australia's best enthusiast mag?

Get your copy today :)
4 Comments
Thoughts on this article? Add a comment below.
iamthemaxx
Oct 7, 2009 12:59 PM
That is for server level environments though, ie 24x7.

Still, it's a lot higher than a lot of people would consider. I can also see how it could be motherboard based.
pappes
Oct 7, 2009 1:38 PM
"The study seems to narrow the problem down to the motherboards that the DRAM is running in rather than the sticks themselves.... standard DIMMs we run don't have even basic error checking."


Tell me how error checking on the DIMMS would prevent corruption from the motherboard?
TheFrunj
Oct 7, 2009 1:58 PM
Uh, it wouldn't. Error checking is for errors in the sticks themselves, so coupled with the errors on the motherboard side the lack of error checking makes it more likely for errors to occur.

-JR
DiStOrTeD
Oct 8, 2009 9:52 PM
Are you suggesting error checks on the data itself or the sticks? You could always take servers down for a memtest run (if you can have scheduled mantainence). If you're talking about live error checking it would decrease response times imo.
Login or register to submit a comment.
 
 
 
Atomic Magazine

Issue: 107 | December, 2009

Atomic is a magazine aimed squarely at computer enthusiasts, gamers, and serious PC upgraders.

Every month we bring you the latest reviews of new technology and PC components, in depth features on everything from overclocking to console hacking, and gaming previews and interviews.
 
Latest Comments
"Fucking signed.

"
by index680i | Nov 21, 2009 2:54 PM
 
""sudo preupgrade"
...failed to download installer metadata
------------
So ..."
by wlayton27 | Nov 21, 2009 8:16 AM
 
"I thought Vista outlived it's usefulness about the same time it was released , lol"
by mr.gargoyle | Nov 21, 2009 12:28 AM
 
"^ I find with CoD4 that I can jump on an empty server and be joined by 6-12 others before the ..."
by Ezekill | Nov 20, 2009 10:10 PM
 
"check

LOMAC
DCS Black Shark
X-plane"
by Bastard Child | Nov 20, 2009 8:13 PM
Latest User Reviews
Shenmue II
10%
asdfasdf
 
EVGA X58 Classified
90%
great board, a few things could be better
 
EVGA X58 Classified
90%
Gorgeous looking
 
Sapphire 4890
90%
So good, I immediately wanted a second one!
 
MSI 790FX-GD70 motherboard
90%
Allmost the prefect gaming board